Chapter III. Research Methodology
3.5 Data Analysis
3.5.1 PLS-SEM
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
- 32 -
3.5 Data Analysis
To validate the research framework, I conducted a survey to test the hypotheses. I analyzed the data collected using SmartPLS to apply variance-based structural equation model (PLS-SEM).
3.5.1 PLS-SEM
Hair et al (2011) had suggest this method to be known as silver bullet since there are a lot of advantages comparing to covariance-based structural equation model (CB-SEM). For example, PLS-SEM works better at maximizing the explain variance of latent constructs.
Afthanorhan (2013) also have proven that the confirmatory factor analysis (CFA) conducted by PLS-SEM is more reliable and valid. Based on the result section, the value of factor loadings/outer loadings, and average variance extracted (AVE) in PLS-SEM is better than CB-SEM even when using the same data provided.
The reason why I chose PLS-SEM for analyzing data was also because it makes analyzing abstract concept possible. Taking advantage of PLS-SEM, I got to analyze vague and complicated concept such as involvement. In PLS-SEM, abstract concepts are constructs which contain several variables to explain the idea. Figure 16 illustrates the procedure for executing PLS-SEM.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 16: PLS-SEM Procedure 3.5.2 Model Verification
The model verification includes two parts: confirmatory factor analysis (CFA) and hypothesis testing. CFA includes the assessments of goodness-of- fit, composite reliability, and construct validity of the model proposed in the study. Constructive validity includes convergent validity and discriminant validity.
To achieve convergence validity, the factor loading of each variables should be larger than 0.5 (Hair et al, 2010), average variance extracted (AVE) should be larger 0.5, composite reliability and Cronbach’s alpha should be at least larger than 0.6 (Hair et al, 2010). To achieve discriminant validity, the root if average variance extracted (AVE) should be larger than the correlations of latent variables. The goodness-of-fit could be calculated with the average of AVE and the average of R square.
Interpretation Model Verification
Data Collection
Constructing Model and Choosing Measure Tool Literature Review
Defining Research Question and Research Design
‧
samples were considered invalid due to that the participants were not LINE user or the answers given by the participants were making no sense. Therefore, there were 83 valid samples.3.6.1 Demographic analysis for Pretest
Among these 83 samples, there are 42 female partic ipants and 41 male participants, so the gender of the sample is quite equal. Most of the participants of the pretest are 21 to 30 years old. The education level of most of the participants of pretest is college and most of them are student.
The monthly income of most of the participant of pretest is below 10,000NTD. The overall demographic data of the participants o f pretest are showed in table 10.
Table 10: Demographic analysis of pretest sample
Category Item
Education Postgraduate 24 29%
‧
‧
3.6.2 Measurement Assessment for Pretest
The overview of quality criteria of the pretest and the factor loading of each variable is shown in table 11. As shown in Table 12, all composite reliability and Cranbach’s Alpha are larger than 0.7. Also, most of the AVE is larger than 0.5, except for the AVE of the construct, Network Effects, is only 0.41 which indicate that the construct is not well explained by the variables.
Looking at the factor loading, we can see the first and second question of the construct might be the one causing problem. Furthermore, the values of factor loading should be larger than 0.5, N E1, NE2, and I4 are only 0.45, 0.3 and 0.43. Therefore, I decide to delete these three questions from the questionnaire. The goodness-of-fit of the model is 0.51 which is acceptable.
Therefore, reconstructing the model is not necessary. I will only adjust the variables.
‧
‧
discarded due to that the participants answering the questionnaire are not LINE users or the answers provided are meaningless. Therefore, there were 301 units of valid samples.In the formal questionnaire, questions were grouped into five sections. In each section, there was a short description to explain and lead the participant into a scenario. The five sections contained general questions about user, user’s experience, experience regarding in- App purchases, user’s perspective toward in-App purchases and user’s ability regarding purchase in LINE. The matching of the constructs and the sections are showed in table 12.
Table 12: Questionnaire Sections and Constructs
Section Construct No. of Questions
General Questions 6
User’s Experience Perceived Playfulness 7 Purchase Experience Social Network Effects 5
User’s perspectives
Personal Involvement 8
Attitude toward In-App Purchasing
2
User’s Ability
Attitude toward In-App Purchasing
1
Intention toward In-App Purchasing
4
‧
Personal Information 6
4.1 Demographic analysis
shows the demographic analysis of the samples.Table 13: Demographic Analysis
Category Item
Postgraduate 119 39.5%
Undergraduate 161 53.5%
Senior High School 8 2.7%
‧
Military/Police 10 3.3%
Architect 9 3.0%
10,001-20,000NTD 38 12.6%
20,001-30,000NTD 39 13.0%
30,001-40,000NTD 53 17.6%
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
40,001-50,000NTD 27 9.0%
Over 50,001NTD 33 11.0%
From the demographic analysis, I found some differences between genders.
According to the analysis, female participants are more indulged in LIN E since they tend to forget time and what they should be doing. Male participants tend to be effected by social network more than females. However, in the construct of network effects, females tend to be influenced by commercials or advertisements more than males. Regarding the intention to purchase in LINE, males have higher intention in purchasing than females and have a better understanding on how to purchase in LINE. Table 14 shows the comparison between genders.
Table 14: Comparison between Genders on Average of each Construct
PP NE PI A I
Male 3.60 3.89 3.85 4.05 4.23
Female 3.92 3.56 3.76 3.97 3.99
Difference -0.32 0.33 0.09 0.08 0.24
*PP stands for Perceived Playfulness; NE for Network Effects; PI for Personal Involvement; A for Attitude toward in-App Purchasing; I for Intention toward in-App Purchasing.
In addition to the difference between genders, I also compare iOS and Android users. Although these two types of user show no significant difference in terms of user’s experience, they are difference in terms of the attitude toward in-App purchases and the intention toward in-App purchases. IOS users show higher attitude and intention toward purchasing in LINE which means that iOS users think that purchasing in LIN E is a good idea and they tend to purchase
‧
Table 15: Comparison between users on Average of each Construct
PP NE PI A I
iOS 3.70 3.73 3.88 4.13 4.27
Android 3.80 3.74 3.78 3.95 4.03
Difference -0.09 -0.01 0.09 0.18 0.24
*PP stands for Perceived Playfulness; NE for Network Effects; PI for Personal Involvement; A for Attitude toward in-App Purchasing; I for Intention toward in-App Purchasing.
4.2 Confirmatory Factor Analysis
Before testing the hypothesis, I will evaluate the reliability and validity of the model first. See table 16 for the overview of CFA.
Table 16: Overview of CFA
Construct Item Cronbach’sα
Composite Personal Involvement; A for Attitude toward in-App Purchasing; I for Intention toward in-App Purchasing.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4.2.1 Reliability Test
In general, if the value of Cronbach’s α is larger than 0.7, then value of composite reliability is larger than 0.6 and the value of AVE is larger than 0.5, we consider that the measurements are consistent and stable which means the questions are with well reliability (Bagozzi & Yi, 1988; Fornell &
Larcker, 1981). As shown in table 16, all Cronbach’s α is larger than 0.7, all composite reliability is larger than 0.6 and the value of AVE is larger than 0.5, therefore, the constructs are reliable.
4.2.2 Validity Test
Validity test include convergent validity and discriminant validity.
Convergent validity refers to that the items in the same construct are related to each other. Discriminant validity refers to that the items from different constructs are in fact different than each other.
To examine the convergent validity, we look at Cronbach’s α and composite reliability. These two values have to be over 0.7. As shown in table 16, the standard is achieved. As for discriminant validity we have to compare the factor loading with cross- loading, and the root of AVE with the correlation between each construct. The factor loading of the construct should be larger than cross- loadings. The root if AVE should be larger than the correlation between each constructs. See table 17 for the matrix of loadings and table 18 for the comparison of root of AVE and correlations.
As shown in table 17, all factor loadings are larger than cross- loadings.
In table 18, all roots of AVE are larger than the correlation with other constructs. Thus, the validity assessment is achieved.
‧
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
PP4 0.20 0.12 0.33 0.24 0.69
PP5 0.50 0.43 0.36 0.51 0.82
PP6 0.53 0.46 0.31 0.52 0.80
*PP stands for Perceived Playfulness; NE for Social Network Effects; PI for Personal Involvement; A for Attitude toward in-App Purchasing; I for Intention toward in-App Purchasing.
* The colored areas indicate the factor loadings of the construct.
Table 18: Comparison between Root of AVE and Correlations
A I NE PI PP
A [0.95]
I 0.68 [0.75]
NE 0.27 0.20 [0.74]
PI 0.84 0.67 0.37 [0.82]
PP 0.45 0.37 0.42 0.48 [0.73]
*PP stands for Perceived Playfulness; NE for Social Network Effects; PI for Personal Involvement; A for Attitude toward in-App Purchasing; I for Intention toward in-App Purchasing.
*The values in the brackets indicate root of AVE of the construct.
4.3 Structural Model Assessment
Structural model examined the relationships between different constructs, hence, from the analysis I tested if the significance of the hypotheses. The analysis was done by SmartPLS applying regression and bootstrapping. Figure 17 shows the path analysis output of the model.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
- 46 -
Figure 17: Path Analysis
Path analysis shows the relationships and intensity between each constructs.
The values of the coefficients are between 1 and -1. If the coefficient is positive, it means that two construct are related positively. If the coefficient is negative, it indicates that two constructs have an inverse relationship. The t-value and the path coefficient are shown in table 19.
Table 19: t-Value and Path Coefficient
Construct
Path Coefficient
t-Value p-Value Vitrified
NE → PP 0.420 9.15 <0.001 Significant
NE → PI 0.467 7.58 <0.001 Significant
PP → A 0.079 2.29 <0.05 Significant
NE → A -0.064 1.89 <0.1 Not Significant
PI → A 0.824 30.27 <0.001 Significant
A → I 0.684 21.44 <0.001 Significant
Since SmartPLS did not provide p-values, I estimated it using two-tail test.
As shown in table 19, all relationships except for the relationship of “Social Network Effects toward Attitude” have a t-value larger than 1.96 which indicates
‧
According to the path analysis and t-value we conclude that all hypotheses except for H4 are significant. See table 21 for hypotheses testing summary.
Table 20: Summary of Hypotheses Test Result
Hypothesis t-Value p-value Support H1: Social network effects are positively
associated with perceived playfulness of the App
9.15 <0.001 Yes
H2: Social network effects are positively associated with Personal Involvement.
7.58 <0.001 Yes
H3: Perceived playfulness is positive associated with attitude toward in- App purchase
2.29 <0.05 Yes
H4: Social network effects are positively associated with attitude toward in-Apps purchasing
1.89 <0.1 No
H5: Personal involvement is positively associated with the attitude toward download and use.
30.27 <0.001 Yes
H6: Attitude toward in-App purchasing is positively associated with intention to purchase.
21.44 <0.001 Yes
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
- 48 -
Chapter V. Discussion
5.1 Discussion
The prevailing state of smart handsets also stimulates the development of App industry. Among all profit models of Apps, in- App purchasing is becoming the favorable one. This study strives to understand the causes that would influence users’ behavior when deciding to purchase in-App products.
5.1.1 Summary of Results
The result of current study shows that social network effects have an impact on users’ perceived playfulness and personal involvement. Social network has effect on both perceived playfulness and personal involvement;
however, it influences perceived playfulness more than it does on personal involvement.
We also found that personal involvement and perceived playfulness have an impact on attitude toward in- App purchases. Personal involvement is the strongest predictor of attitude toward in-App purchase. Both perceived playfulness and personal involvement plays a role in influencing the attitude, but personal involvement plays a more important role than perceived playfulness. Surprisingly, social network effects do not have a direct impact on attitude toward in-App purchases. The reason could be that people nowadays have higher ego consciousness and the information is more transparent than before as a consequence of the prevailing the internet.
Also, attitude toward in-App purchases have an impact on intention to purchase in- App product. This hypothesis is supported just as it is in technology acceptance model (TAM).
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 18 shows the model indicates by the result of the study. Social network effects do not have a direct impact on attitude toward in-App purchasing, therefore, personal involvement and perceived playfulness become the mediators between social network effects and attitude toward in-App purchasing.
Figure 18: Model indicated by Research Result 5.1.2 Contribution and Key Insights
The primary contribution of this study is that we found social network effects do not have a direct impact on attitude toward in-App purchases which is very different from the research outcome of our previous study, viz in A Research into the Diffusion Effects of Free-Trial to Mobile Applications (2013). The result of afore mentioned shows, social network effects greatly affect users’ attitude toward download and trial.
According to Rogers (1983), the diffusion of an innovation is directly affected by social networks, seems like it might not be true anymore owing to the fact that people nowadays have higher ego consciousness and that they can reach information about the innovation more easily than before.
Whenever people come across information regarding an in- App product, they will try to understand it more before they make decision to purchase it. This behavior indicates that when making decision to purchase in-APP products,
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
- 50 -
people’s mind are following the “central route” defined in elaboration likelihood model (ELM).
5.2 Research Limitations
This research is limited to communication Apps which is the category LIN E represents, in order to have a wider understanding of in-app purchasing, a wider variety of users would be more appropriate. However, users’ behavior across different categories of Apps might be different which will make this research too complicated to conduct in this short period of time.
Furthermore, since the research samples are collected mainly on campus, the participants are mainly at age of 21 to 30 and most of them are students. This might lead to generalized validity problem which means the study outcome might not be able to stand for users of all age but limited to users that match the condition of the study.
5.3 Conclusion
In conclusion, this study was conducted to examine factors influencing behavioral intention in purchasing in-App products. Our research model and hypothesis were based on Innovation Diffusion Theory, Extended Technology Acceptance Model, and Elaboration Likelihood Model. We surveyed LINE users, mostly aged 21 to 30, and found support for five of six hypotheses. The result of this study confirmed the important roles of attitude, perceived playfulness and personal involvement in predicting behavioral intention, and pointed out that social network effects have no direct impact on intention to purchase in-App products.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5.4 Future Research and Suggestions
Future research can try to discuss the value created directly from APP developers, and try to analyze APPs in different categories. As mentioned in research limitation, owing to time limited, we narrowed down the scope of the study.
The scope of the study only covered users’ view toward the product within Apps includes in the category of communication, but neglected the part regarding the interactions between users and App developers. The value created during the interactions which is the part that can be discussed using value creation cycle (Yu, 2012) is also noteworthy.
Furthermore, users might have different thinking when it co mes to different category of Apps. Apps from different category might have different characteristics. This study focus on communication Apps, and took LIN E as an example. The characteristics pointed out in this study focused are mainly of the category.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
- 52 -
Reference
Afthanorhan, W. M. A. B. W. (2013). A Comparison Of Partial Least Square Structural Equation Modeling (PLS-SEM) and Covariance Based Structural Equation Modeling (CB-SEM) for Confirmatory Factor Analysis. International Journal Engineering and Science Innovative Technologies (IJESIT), 2(5), 8.
App Annie and International Data Cooperation (2014). Mobile App Advertising and Monetization Trends 2012-2017: The Economics of Free. Retried from App Annie website: http://goo.gl/QPMgS3
Bagozzi, R. P., & Yi, Y. (1988). On the evaluation structural equation models.
Academic of Marketing Science, 16, 74-94
Bagozzi, R. P., Davis, F. D., Warshaw, P. R. (1992), Development and test of a theory of technological learning and usage. Human Relations, 45(7), 660–686.
Bhattacherjee, A., Hikmet, N. (2007). Physicians’ Resistance toward Healthcare Information Technology: A theoretical Model and Empirical Test. European Journal of Information Systems, 16 (6), 725-737.
Booth-Butterfield, S., Welbourne, J. (2002). The Elaboration Likelihood Model: Its Impact on Persuasion Theory and Research. In J. P. Dillard & M. Pfau (eds.), the Persuasion Handbook: Developments in Theory and Practice. Thousand Oaks, CA: Sage, pp. 155-173
Brancheau, J. C., Wetherbe, J. C. (1990). The Adoption of Spreadsheet Software:
Testing Innovation Diffusion Theory in the Context of End-User Computing.
Information Systems Research. 1(2), 115-143.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Bruce, K. (2013). Use of mobile technology for monitoring and evaluation in international health and development programs [Abstract]. The University of North Carolina at Chapel Hill, California. Retrieved from ProQuest
Dissertations & Theses.
Chau, P.Y., Hu, P.J.H. (2001) Information Technology Acceptance by Individual Professionals: A Model Comparison Approach. Decision Sciences, 32(4), 699-719.
Chiang, J. K., Hung, K.L. & Huang, S.Y. (2013, July). A Research into the Diffusion Effects of Free-Trial to Mobile Applications. Business Intelligence for
Innovation, 國立政治大學商學院服務創新研究中心與資管系.
Cooper, D. R., & Schindler, P. S. (2008). Business Research Methods 10th ed.
New York, NY: McGraw-Hill
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340.
Flynn L.R., Goldsmith R.E. (1993). Application of the personal involvement inventory in marketing. Psychology & Marketing, 10(4), 357-366.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(3), 39-50.
Forrest, C. (2014, March 21). Apple v. Google: The goliath deathmatch by the numbers in 2014. Tech Republic. Retrieved May 19, 2014, from
http://www.techrepublic.com/article/apple-v- google-the- goliath-deathmatch-by-the-numbers- in-2014/
Gartner Inc. (2014). Market Share Analysis: Mobile Phones, Worldwide, 4Q13 and 2013. Retrieved from Gartner website: https://www.gartner.com/doc/2665319
‧
The Journal of Marketing Theory and Practice, 19(2), 139-152.
Heggestuen, J. (2013, December 19). Global Messaging App LINE Sees Explosive Revenue Growth from Stickers, Games, and Paid Accounts. Business Insider.
Retrieved May 5, 2014, from Web-based Learning Communities. National Kaohsiung Normal University, Kaohsiung. Doctoral desertion retrieved from National Digital Library of Thesis and Dissertations in Taiwan.
Hu, P.J., Chau, P.Y., Sheng, O.R.L., Tim, K.Y. (1999). Examining the Technology Acceptance of Telemedicine Technology. Journal of Management Information Systems, 16(2), 91-112.
Huang, L. Y., & Hsieh, Y. J. (2011). “Predicting online game loyalty based on need gratification and experiential motives.” Internet Research, 21(5), 581-598.
Ingraham, N. (2013, October 22). Apple Announces 1 Million Apps in the App Store, More Than 1 Billion Songs Played On iTunes Radio. The Verge. Retrieved March 8, 2014, from
http://www.theverge.com/2013/10/22/4866302/apple-announces-1- million-apps -in-the-app-store
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Kim, D., Chang, H. (2007). Key Functional Characteristics in Designing and
Operating Health Information Websites for User Satisfaction: An Application of the Extended Technology Acceptance Model. International Journal of Medical Informatics, 76 (11-12), 790-800.
Klein, R. (2007). An Empirical Examination of Patient-physician Portal Acceptance.
European Journal of Information Systems, 16(6), 751-760.
Koetsier, J. (2014, March 27). Mobile app monetization: Freemium is king, but in-App ads are growing fast. VentureBeat. Retrieved June 19, 2014, from http://venturebeat.com/2014/03/27/mobile-app-monetization- freemium-is-king-but-in-App-ads-are-growing- fast/
Koetsier, J. (2014, March 27). Mobile app monetization: Freemium is king, but in-App ads are growing fast. VentureBeat. Retrieved June 19, 2014, from http://venturebeat.com/2014/03/27/mobile-app-monetization- freemium-is-king-but-in-App-ads-are-growing- fast/